Constrained Molecule Generation Modelled Using the Grammar Constraint

Drug discovery is a very time-consuming and costly endeavour due to its huge design space and to the lengthy and failurefraught process of bringing a product to market. Automating the generation of candidate molecules exhibiting some of the desired properties can help. Among the standard formats to encode molecules, SMILES is a widespread string representation. We propose a constraint programming model showcasing the grammar constraint to express the design space of organic molecules using the SMILES notation. We show how some common physicochemical properties -such as molecular weight and lipophilicity -and structural features can be expressed as constraints in the model. We also contribute a weighted counting algorithm for the grammar constraint, allowing us to use a belief propagation heuristic to guide the generation. Our experiments indicate that such a heuristic is key to driving the search towards desired molecules.

Code -github.com/cravethedave/MiniCPBP/tree/AAAI26 1 Introduction Drug discovery is a very time-consuming and costly endeavour due to its huge design space -estimated to contain between 10 23 and 10 60 different molecules (Polishchuk, Madzhidov, and Varnek 2013) -and to the lengthy and failure-fraught process of bringing a product to market. Automated molecule design is nowadays a vital part of drug discovery and material science, with computational approaches coming from deep generative models and combinatorial search methods (Du et al. 2022). It aims to extract from this huge design space the most likely candidates according to some desired properties. And even among these, only a few may lead to a usable product after extensive testing.

SMILES, a one-dimensional encoding of molecules, is one of the standards commonly used by this research community. It lends itself well to techniques used for natural language processing, such as sequential generative neural models, but also to constraint programming (CP). Using a context-free grammar and a few additional constraints, we show how to describe valid SMILES strings in a CP model. This allows us to explore the huge design space of possible